Number Reading in Pure Alexiaâ

Home , Aphasia, Pure alexia

Neuropsychologia 49 (2011) 2283–2298

Contents lists available at ScienceDirect

Neuropsychologia

jo urnal homepage: www.elsevier.com/locate/neuropsychologia

Reviews and perspectives

Number reading in pure alexia—A review

a,∗ b

Randi Starrfelt , Marlene Behrmann

Center for Visual Cognition, Department of Psychology, Copenhagen University, O. Farimagsgade 2A, DK-1353 Copenhagen K, Denmark

Department of Psychology, Carnegie Mellon University, Pittsburgh, PA, USA

a r t i c l e i n f o

a b s t r a c t

Article history: It is commonly assumed that number reading can be intact in patients with pure alexia, and that this

Received 25 October 2010

dissociation between letter/word recognition and number reading strongly constrains theories of visual

Received in revised form 31 March 2011

word processing. A truly selective deﬁcit in letter/word processing would strongly support the hypothesis

Accepted 22 April 2011

that there is a specialized system or area dedicated to the processing of written words. To date, however,

Available online 4 May 2011

there has not been a systematic review of studies investigating number reading in pure alexia and so

the status of this assumed dissociation is unclear. We review the literature on pure alexia from 1892 to

Keywords:

2010, and ﬁnd no well-documented classical dissociation between intact number reading and impaired

Pure alexia

letter identiﬁcation in a patient with pure alexia. A few studies report strong dissociations, with number

Alexia without agraphia

reading less impaired than letter reading, but when we apply rigorous statistical criteria to evaluate

Letter-by-letter reading

Visual recognition these dissociations, the difference in performance across domains is not statistically signiﬁcant. There

Number reading is a trend in many cases of pure alexia, however, for number reading to be less affected than letter

Dissociation identiﬁcation and word reading. We shed new light on this asymmetry by showing that, under conditions

of brief exposure, normal participants are also better at identifying digits than letters. We suggest that

the difference observed in some pure alexic patients may possibly reﬂect an ampliﬁcation of this normal

difference in the processing of letters and digits, and we relate this asymmetry to intrinsic differences

Contents

1. Introduction ...... 2284

2. Literature review ...... 2284

2.1. Methods ...... 2284

2.2. Results ...... 2285

2.2.1. No dissociation ...... 2285

2.2.2. Possible dissociations ...... 2286

2.3. A new analysis of published data ...... 2288

2.4. Summary of results ...... 2289

3. Normal performance and visual similarity ...... 2289

3.1. Normal recognition of letters and digits ...... 2289

3.1.1. Methods...... 2289

3.1.2. Results ...... 2290

3.2. Visual similarity for letters and digits...... 2290

3.2.1. Methods...... 2290

3.2.2. Results ...... 2291

4. Discussion ...... 2291

Acknowledgements ...... 2293

Appendix A...... 2293

References ...... 2297

∗

Corresponding author. Tel.: +45 35324886.

E-mail address: [email protected] (R. Starrfelt).

2284 R. Starrfelt, M. Behrmann / Neuropsychologia 49 (2011) 2283–2298

1. Introduction patient’s score on task I is markedly lower than on task II, but where

performance is not compared to a control group; (ii) a strong disso-

Pure alexia is an acquired disorder of reading that leaves writ- ciation, where “neither task is performed at a normal level, but task

ing and other language skills unaffected (hence the term ‘pure’). I is performed very much better than task II” (p. 228). In this case,

The disorder is characterized by slow and effortful but mostly cor- performance is commonly compared to a normal control group,

rect reading, with a pronounced effect of word length on reading but a patient’s performance on tasks where normal controls would

latency: Reaction times increase linearly with word length, often be expected to perform at ceiling may also constitute evidence

with hundreds of milliseconds per letter. Pure alexia is associated for a strong dissociation (i.e., in the absence of directly compar-

with a lesion impacting occipito-temporal areas of the left hemi- ing the patient to a control, one assumes perfect or near-perfect

sphere, and damage to the left mid-fusiform gyrus, in particular, performance for normal individuals); and (iii) a classical dissocia-

seems to play a causal role in this disorder (Leff, Spitsyna, Plant, tion, where – relative to normal controls – performance on task I

& Wise, 2006). Other forms of acquired reading disorders, com- is impaired while performance on task II is within normal limits.

monly termed central alexias, are seen following more widespread While trend dissociations are taken as a weak form of evidence,

damage to temporo-parietal areas in the left hemisphere, and are both strong and classical dissociations have been interpreted as

associated with more general deﬁcits in language like aphasia and suggestive of specialized functions or modularity. Recently, more

agraphia. A question that continues to plague theories of pure alexia reﬁned (and operational) criteria specifying statistical demands for

is whether the disorder is speciﬁc to alphabetic characters, and is dissociations have been suggested (e.g., Crawford & Garthwaite,

caused by damage to or disconnection of a cerebral area dedicated 2005, 2007; Crawford, Garthwaite, & Grey, 2003). In brief, this

to visual letter or word recognition (the so-called Visual Word Form approach demands that, in order to conclude that there is a classi-

Area; Cohen et al., 2003, 2004; Gaillard et al., 2006), or whether it cal dissociation, the patient’s scores should differ signiﬁcantly from

emerges from a more general visuoperceptual deﬁcit (Behrmann, the control group on one of two tasks, while performance should

Nelson, & Sekuler, 1998; Farah & Wallace, 1991; Starrfelt, Habekost, be within the normal range on the other task, and – importantly

& Leff, 2009). – the difference between the patient’s standardized scores in the

One way to adjudicate between these alternatives is to examine two tasks should be signiﬁcant. For the less stringent strong dissoci-

the performance of pure alexic patients when presented with other ation, there should be a signiﬁcant difference between the patient’s

visual stimuli. It has often been suggested that number reading can standardized scores on the two tasks in question, while both scores

be intact in patients with pure alexia (e.g.,Dehaene, 2009; Leff et al., may differ signiﬁcantly from the mean of the control group.

2001), but, to date, there has not been a systematic review of the The presence of a clear classical dissociation would provide

evidence regarding number reading in these patients. If number strong theoretical support for the independence or segregation of

reading can remain intact following a lesion causing pure alexia, the system subserving processing of letters or words from the sys-

this would offer convincing evidence of the speciﬁcity and selec- tem(s) subserving the visual processing of numbers. The absence

tivity of the disorder, and would support the hypothesis that pure of a classical dissociation but evidence for a strong dissociation or

alexia affects the processing of alphabetic material only. If, on the a trend dissociation would require further explication—numbers

other hand, number reading is invariably impaired in patients with and letters/words might be mediated by separate systems, both

pure alexia, this would suggest a more general (visual) deﬁcit being of which are damaged albeit differentially, or the two domains

at the core of the disorder. In this paper, we review the literature might potentially be subserved by the same system. If the latter

on pure alexia from 1892 until 2010, and select for further inves- were true, an explanation that accounts for the differential impair-

tigation all papers that describe a pure alexic patient’s reading of ment between the two domains will be required. Clear evidence

numbers. Our aim is to characterize the relationship between the for separated areas or modules responsible for letter and digit pro-

reading of letters/words and numbers in pure alexia, and, in partic- cessing must be supported by a pattern of double dissociation in

ular, to search for a possible dissociation of performance between which the complementary pattern of impaired and preserved abil-

these two symbol types. ities is observed. Thus, if we ﬁnd a clear dissociation between letter

An immediate challenge concerns the demarcation of pure and number reading in a patient with pure alexia, searching for a

alexia: although the deﬁnition of pure alexia is quite straightfor- patient showing the opposite pattern (number reading impaired,

ward, different labels have been used for this disorder over the letters spared) would be the obvious and necessary next step.

last century or so. Thus, all the following terms have been used In sum, the goal of our investigation is to determine whether a

to describe the same, or very similar, disorders: pure alexia, global dissociation between the reading of numbers and letters/words can

alexia, alexia without agraphia, visual alexia, verbal alexia, word blind- be found in pure alexia, and if so, to examine the strength of the dis-

ness, and letter-by-letter (LBL) reading. Some of the labels describe sociation using Shallice’s classiﬁcations and the operational criteria

a variation in severity, for instance ‘global alexia’ refers to a total speciﬁed by Crawford et al. (2003), Crawford and Garthwaite (2005,

inability to read letters and words, while ‘pure alexia’ usually 2007). We start by conducting an extensive review of the existing

(but not consistently) refers to patients who are still able to read, literature on pure alexia, focusing on studies reporting results from

although at an abnormal speed (and presumably using an abnormal number reading tasks. We then go on to present new data from a

strategy). Also, many different patients have been grouped under psychophysical study of letter and digit identiﬁcation in normal

the heading of ‘LBL reading’, regardless of whether or not they subjects, and, ﬁnally, we report an analysis of differences in image

are pure alexic. Many patients with LBL-reading have accompa- statistics between the two symbol types.

nying deﬁcits in writing (e.g., Ingles & Eskes, 2008; Lambon Ralph,

Hesketh, & Sage, 2004) and/or naming (e.g., Lambon Ralph et al.,

2004), and/or make reading errors commonly seen in more central 2. Literature review

types of alexia (e.g., Friedman & Hadley, 1992; Ingles & Eskes, 2008).

We have chosen to include all patients with disorders referred to by 2.1. Methods

any of the labels mentioned above, and we comment on this issue

in the results section. For this review of the neuropsychological literature, we have

Another challenge concerns what constitutes a dissociation. As included all studies of patients with pure alexia, global alexia, alexia

clearly laid out by Shallice (1988), there are (at least) three types of without agraphia, visual alexia, verbal alexia, or letter-by-letter (LBL)

dissociations in neuropsychology: (i) A trend dissociation, where a reading. As mentioned above, these labels are often (but not always)

R. Starrfelt, M. Behrmann / Neuropsychologia 49 (2011) 2283–2298 2285

used to refer to the same type of reading disorder, and, to ensure reading). We include in our literature review all patients referred

that no cases were missed, we adopted a policy of overinclusion to by any of these different labels, and ﬁnd no systematic pattern

in this ﬁrst step. A thorough search of Web of Science, PubMed, in the type of diagnosis/label for the patients’ reading deﬁcit and

PsychInfo, and ScienceDirect, as well as The Copenhagen Neu- associated deﬁcits, in relation to patients’ performance with num-

ropsychology Database (www.gade.psy.ku.dk/database), using the bers. We have thus kept all patients in the analysis, although some

above-mentioned keywords was performed. The latest search was have associated deﬁcits like agraphia (and thus do not strictly con-

conducted on August 27th 2010. Of the retrieved references, only form to a diagnosis of alexia without agraphia or pure alexia). In

studies of patients reading a script using Latin letters were included. the discussion of individual patients in the following sections, their

Although interesting, associations and dissociations between read- associated deﬁcits are commented on, and these are also listed in

ing of different Asian orthographies like Japanese Kanji and Kana Table A1.

are difﬁcult to interpret in the current context. Only papers pub-

lished in English were included, with one exception: because of 2.2.1. No dissociation

its importance, we include Dejerine’s (1892) original patient study In 46 of the 90 reviewed patients, no dissociation between per-

(translated in Bub, Arguin, & Lecours, 1993; Rosenﬁeld, 1988). Con- formance with letters and digits is reported. For 36 of these 46

ference abstracts were generally disregarded, but one (Henderson, individuals, both letter reading and number reading are reported

1987) speciﬁcally describing a study of number reading in pure to be impaired (cases 1–36 in Table A1). In many of these cases, the

alexia, was included. Because of our focus on acquired deﬁcits in tasks administered to assess letter reading and number reading are

the mature visual system, studies of children, including reports of not directly comparable, but roughly the same level of impairment

developmental dyslexia, were also excluded. is noted with both kinds of symbols. The majority of these patients

From the assembled list, containing a total of 223 references, we will not be considered further (but see section 3.1. for a discussion of

selected all studies mentioning patients’ performance with digits cases 33–36 reported by Starrfelt et al., 2009). In the remaining ten

or Arabic numerals, whether formally assessed or just mentioned in patients, the processing of (single) letters and of digits is reported to

the descriptive background information. This resulted in 76 papers be unaffected (cases 37–46 in Table A1). For seven of these patients,

reporting a total of 90 different patients, all of whom are included no details of assessment are presented, and it is merely stated that

in this review. An overview of these papers and summary of the the patient is able to name letters and numbers (cases 37, 38, 40,

patient data can be found in Table A1 in Appendix A, which also 42, 43, 44, 45), which makes it difﬁcult to evaluate whether more

lists (as far as possible) diagnosis given, aetiology, lesion site, visual subtle deﬁcits were present in these patients and/or whether any

ﬁeld defect, and whether the patient was reported to have aphasia deﬁcits might have been uncovered with more sensitive testing.

or agraphia. The remaining three patients are presented in some detail below.

Caffara (1987; case 39) reports ﬂawless naming of single letters

2.2. Results with 30 ms tachistoscopic presentation, in a patient with alexia fol-

lowing a left occipito-temporal hematoma. Naming of single digits

Many of the reviewed papers only summarize patient perfor- was without error (12 trials) with 50 ms presentation. The exposure

mance brieﬂy, without presenting details of assessment methods durations were chosen to “obtain the best approximate perfor-

or speciﬁc results. Further, even when details are presented, most mance with the lowest presentation time” (p. 68). No normal data

studies lack a control group against whose performance the patient are presented, so it is unknown whether normal participants would

data can be compared. The absence of the normal benchmark makes be able to report letters or digits at even lower exposure durations,

it difﬁcult to judge whether any observed difference between but at least the patients’ performance on both stimulus types is

patients’ performance with letters and digits reﬂects a normal pat- good at relatively brief presentations times. These tests were per-

tern or asymmetry, or whether it is indicative of a true dissociation. formed three weeks after onset, and this patient’s alexia remitted

Also, in many of the reviewed studies (as in most neuropsychologi- within ten months.

cal case studies in general), tasks are often quite simple in the sense Relatively intact identiﬁcation of letters and numbers was also

that normal subjects would be expected to perform at ceiling. On noted in patient ROC reported by Warrington and Langdon (1994;

the one hand, using simple tasks makes it unlikely that a difference 2002; case 41), who was diagnosed with spelling dyslexia follow-

between the tasks would be observed in the normal population, ing a large occipito-parietal stroke. ROC had a left uniocular nasal

but, on the other hand, we do not know whether a difference in quadrantanopia. ROC’s threshold for single letter identiﬁcation was

performance might be observed in normal participants under more 35 ms, which is at the same level as a non-alexic control patient

difﬁcult tasks (when the ceiling effect is removed). Most normal with spelling problems (Warrington & Langdon, 2002). No normal

subjects do not make errors in naming letters and digits under con- controls were tested, so it remains unknown whether ROC’s perfor-

ditions of free vision, but differences in performance with the two mance is within normal bounds. At the very least, ROC’s single letter

kinds of characters might still be revealed by more sensitive and recognition ability is superior to that commonly observed in pure

taxing measures. alexia. ROC also performed speeded number copying at normal age

On this basis, we have grouped the individual case reports level, relative to control subjects tested in another study. As the

according to two parameters; whether or not they report a dis- two tasks (recognition and copying) are not directly comparable, it

sociation between performance with letters/words and digits, and, is difﬁcult to evaluate whether processing of both kinds of symbols

for those cases where a possible dissociation is present, how well were (un)affected to the same degree, but the presented data at

supported this dissociation is by the evidence at hand. This has least indicates that both letter and number reading was relatively

resulted in the four main groups of patients that will be described normal.

in the following sections (2.2.1–2.2.2.2.). Table A1 in Appendix A Patient FC, reported by Rosazza, Appollonio, Isella, and Shallice

includes a numbered list of patients; these numbers are used for ref- (2007; case 46), suffered from pure alexia and a right homony-

erence in the following sections as some papers report more than mous hemianopia following stroke. His main lesion affected the

one patient, and patients reported in the same paper may show left occipito-temporal region, but occipito-temporal structures in

different patterns of performance. the right hemisphere were also affected, in addition to multiple,

As mentioned in the introduction, pure alexia may be referred to diffuse foci of chronic ischemic encephalopathy. FC had reaction

by different labels, and patients with associated deﬁcits like anomia times (RTs) within the normal range for both letter and digit nam-

or agnosia may be referred to by some of these labels, too (e.g., LBL- ing, and was also within normal limits on letter identiﬁcation under

2286 R. Starrfelt, M. Behrmann / Neuropsychologia 49 (2011) 2283–2298

rapid conditions. No accuracy scores are given for the letter and was impaired not only in letter identiﬁcation, but also in number

digit naming tasks, so we cannot know whether the patient made reading. It is also clear, however, that his ability to recognize digits

more errors than controls. Assuming that accuracy was also within was superior to his letter recognition, as, according to the descrip-

the normal range, this patient seems to have intact letter and digit tion, he was not able to name even a single letter from visual input

identiﬁcation skills. FC’s RTs in word reading were slow (mean RT alone. The information available about Monsieur C’s number read-

was approximately 1500 ms for words of 4–10 letters), with a mod- ing is limited, making it difﬁcult to classify deﬁnitively the type of

est word length effect of 70–90 ms per letter. He was impaired in dissociation present in this case. As he was impaired in both tasks,

some visual tasks (the object decision task from BORB and the sil- and there was no comparison to a control group, we have tenta-

houette task from VOSP). Rosazza et al. (2007) argue that FC has a tively classiﬁed this case as representing a trend dissociation, but

deﬁcit in integrating letters into letter groups and words, and that we note that there seems to be a quite large discrepancy between

this deficit in itself is sufficient to give rise to pure alexia. This case his letter and digit identification skills.

is noteworthy because it may be the only one on record demon- Other studies report a pattern similar to the one observed in

strating normal RTs in both letter and digit naming in pure alexia, Monsieur C. Ajax (1977; case 81) reported a patient who was able

in spite of other visual deﬁcits being present. For our primary pur- to read multidigit numbers up to 10 digits correctly, while he

poses, it is important to note the reported data show no dissociation showed difﬁculties with unisyllabic words, and was unable to read

between performance with letters and digits. The same is true for multisyllablic words. No further details from the reading assess-

the two patients discussed above (Caffarra, 1987; Warrington & ment are presented. In another study, Landis, Regard, and Serrat

Langdon, 1994, 2002). (1980; case 82) reported a patient with alexia without agraphia fol-

lowing a glioblastoma affecting the left occipital lobe. They noted

2.2.2. Possible dissociations that “when presented with combinations of letters and numbers, he

Findings that may indicate a dissociation between performance named the letters slowly but the numbers at normal speed” (p. 48).

with letters and digits are reported in the remaining 44 patients Grossi, Fragassi, Orsini, De Falco, and Sepe (1984; case 83) studied a

(cases 47–90 in Appendix A). For 33 of these, the original papers patient diagnosed with alexia without agraphia, who was nonethe-

lack the details necessary to reach any conclusions about the type less impaired in writing single letters both spontaneously and to

of dissociation, either because tasks with letters and digits are not dictation. This patient could accurately name 30/30 “simple num-

comparable (N = 5; cases 47–51), or, more often, because too few bers” and 16/20 “complex numbers”, while his scores with printed

details are given about assessment methods and stimuli (N = 28; single letters and words were 38/42 and 2/50, respectively. In this

cases 52–79). We note that for all the 33 patients, performance is case, the trend may be weak, as the major difference observed is

reported to be better with digits than letters, a point to which we between ‘complex numbers’ (not further speciﬁed) and words, but

return in the discussion, but these patients will not be considered as there is an indication of dissociation of performance, we have

in further detail here. Six patients (cases 80–85, discussed in sec- included the study here.

tion 2.2.2.1.) fulﬁl the criteria for a trend dissociation, with letter A quite clear dissociation between number and letter read-

reading being more severely impaired than number reading. These ing was reported in the pure alexic patient VT (case 84; Maher,

studies lack a control group or use rather coarse tasks, and statis- Clayton, Barrett, Schober Peterson, & Gonzalez Rothi, 1998). VT

tical comparison between tasks is not conducted. Four papers (ﬁve scored within the normal range on tests of visual perception (the

patients; cases 86–90, section 2.2.2.2.) report a strong dissociation, VOSP or MVPT-V batteries), and scored 59/60 on the Boston nam-

with the patients being impaired with both letters and digits, but ing task. She was able to match physically identical words, but was

disproportionately affected with letters. Two of these papers relate unable to match letters or words to stimuli presented in a different

performance to that of normal controls, but as the patients do not case or font. No single letter naming data are presented, but it is

perform within the normal range with digits, criteria for a classical stated that VT “was not able to name individual letters from visual

dissociation (numbers spared, letters impaired) are not met. input alone” (p. 641), while she could read aloud numbers with

one to ﬁve digits “accurately and without hesitation” (p. 639). In a

search task, VT was able to detect the presence of an X in a string

2.2.2.1. Trend dissociations. Dejerine’s (1892; case 80) original

of letters, but RTs were slow and increased when the X appeared

patient, Oscar C, is commonly described as having a massive pure

towards the end of the string. When searching for an X or a 0 in a

(or global) alexia with spared number reading (e.g.,Dehaene, 2009;

string of numbers, VT’s performance was “rapid and ﬂawless, and

Geschwind, 1966). However, it is clear from the translation of

there was no indication of an effect of place” (Maher et al., 1998;

Dejerine’s work, and, in particular, from his quotes from the oph-

p. 639). Thus, it seems that the patient was better able to perceive

thalmological evaluation by Landolt (Rosenﬁeld, 1988; see also Bub

several digits in a glance, but had to search letter sequences seri-

et al., 1993) that Monsieur C’s number reading was far from perfect.

ally. In this case, the difference in performance with letters and

Landolt noted that “if one shows him numbers, he is able to distin-

numbers seems quite convincing at ﬁrst glance, but unfortunately

guish them, after some hesitation, from the letters” (Rosenﬁeld,

no details of assessment procedure (instructions, stimuli, order of

1988; p. 34). He further reported that Monsieur C “reads the num-

tasks, etc.), response times or normal data are presented, which

bers poorly, since he cannot recognize the value of several numbers

makes it impossible to judge fully the (ab)normality of VT’s perfor-

at once. When shown the number 112, he says, “It is a 1, a 1, and a

mance.

2”, and only when he writes the number can he say “one hundred

VT’s inability to name single letters resembles the performance

and twelve” (p. 35). This latter strategy resembles the letter-by-

of Monsieur C (Dejerine, 1892; case 80), who, by today’s criteria,

letter method employed by many pure alexic patients in word

would be classiﬁed as having global alexia (Binder & Mohr, 1992).

reading in which they spell out the individual letters sequentially

In both patients, number naming seems better preserved than let-

and then assemble the pronunciation from this sequence (“C” “A”

ter naming. A similar pattern is also evident in the global alexic

“T” and then ‘cat’). There seems to be no doubt that Monsieur C

patient (EA) reported by Larsen, Baynes, and Swick (2004; case 85).

EA could name only 24% (6/25) of single letters at 500 ms expo-

sure, while she named 84% (21/25) of single digits correctly. She

We are concerned by this patient’s multiple lesions, and by his modest word

was unable to read three letter common words presented for 2 s.

length effect, which is lower than most pure alexic patients, and closer to the level

She could name 60% (N not given) of two-digit numbers at 500 ms

commonly reported in hemianopic alexia (Leff et al., 2001). Future studies replicat-

ing this ﬁnding in other patients would be highly informative. exposure, while performance deteriorated to 5% correct (1/20) with

R. Starrfelt, M. Behrmann / Neuropsychologia 49 (2011) 2283–2298 2287

Table 1

Statistical comparison of processing speed (C) for single letters and single digits for (a) Patient NN and controls (N = 5; Starrfelt et al., 2010), and (b) Three patients and controls

(N = 9) from Starrfelt et al. (2009). * marks performance signiﬁcantly reduced compared to controls (p < 0.05; Crawford & Howell’s test).

(a) NN Control mean (SD)

Single letter C 22* 143 (47)

◦ a

Single digit C 42 138 (48)

(b) TJ JT JH Control mean (SD)

Single letter C 31* 27* 22* 117 (23)

Single digit C 44* 47* 79* 119 (16)

◦

p = .07, Crawford & Howell’s test.

Correlation between scores in the control group = 0.915.

Correlation between scores in the control group = −0.130.

four-digit numbers. In sum, EA was obviously impaired in read- perfect. Rather, the patients performed at the same level in the

ing numbers, but at least naming of single digits seems relatively ‘reading for comparison’ task as they did in the single digit iden-

preserved compared to single letters. tiﬁcation task. It seems then, that a general deﬁcit in digit reading

In general, number reading seems relatively less impaired com- may be present in all conditions in Cohen and Dehaene’s (1995)

pared with letter reading, in all the six patients mentioned above. study, and that some additional factor (perhaps level of difﬁculty)

However, the tasks used are not very sensitive and most compar- contributes to the patients’ error scores in the addition task com-

isons are based on accuracy scores, which, while useful, might not pared to the more straightforward comparison task and the simple

be sufﬁciently discriminatory. Also, performance is not compared reading task. For the present purposes, the most important ﬁnding

to normal or patient (other lesion site) controls. Thus, although is that both patients were signiﬁcantly better at reading digits than

these studies suggest that number reading may be less affected letters, although their performance with digits was also impaired.

than letter identification in these case of pure alexia, no definitive Whether or not this finding constitutes a strong dissociation in a

evidence is presented. strict sense (Crawford & Garthwaite, 2005) remains unknown, as

performance was not compared to a control group. In addition,

2.2.2.2. Strong dissociations. Very few studies of pure alexic Cohen and Dehaene (1995) report that their patients’ performance

patients have explicitly aimed to compare performance with let- was far better in comparison tasks than in reading out loud: their

ters and digits, and few make formal comparisons of patients’ patients could often decide which of two numbers was the larger,

performance with the two types of symbols. We have identiﬁed even when they misnamed the very same digits. They interpret this

four studies (reporting ﬁve patients, cases 86–90 Table A1) that do as reﬂecting the processing abilities of a (spared) right hemisphere

make such a comparison and ﬁnd a signiﬁcant difference between system capable of identifying (but not naming) single digits, and of

patients’ performance with letters and digits. Only two of these accessing magnitude information. As accuracy was the only depen-

studies compare patient performance to normal controls (Perri, dent measure in this comparison tasks, it remains unknown if the

Bartolomeo, & Silveri, 1996; Starrfelt, Habekost, & Gerlach, 2010), patients’ performance was at a normal level.

while one includes a group of patient controls (Ingles & Eskes, Perri et al. (1996) report a LBL-reader, SP, who, in addition to

2008). The fourth study (Cohen & Dehaene, 1995) may only meet the impaired reading, also showed signs of agraphia and anomia

the formal criteria for a trend dissociation as they do not compare (case 88 in Table A1). SP, unlike most pure alexics, made many

the patients’ performance to controls. We have chosen to include errors in word reading, mostly due to letter confusions or misiden-

the study here, as it had as its stated goal the investigation of num- tiﬁcations. SP also made errors in letter matching and pointing

ber processing skills in pure alexia, and the patients’ performance tasks. He identiﬁed 90% of single digits correctly, while he was 70

with letters and digits in the same task was statistically compared. and 80% correct with lower and upper case letters respectively.

Also, the authors mostly used tasks where normal participants No statistical tests on these data were presented in the paper, but

would be expected to perform at ceiling. We consider these four a comparison of SP’s performance with letters and digits using a

studies in greater detail below. chi-square test (data from Table 1 and text, Perri et al., 1996; p.

In a comprehensive study of two pure alexic patients, Cohen and 394) reveals a signiﬁcant difference in performance with lower

Dehaene (1995; cases 86–87 in Table A1) showed that both patients case letters vs. digits ( = 8.351, p = 0.004), while the difference

made signiﬁcantly more errors in naming single letters than single in performance with upper case letters vs. digits is only borderline

digits when compared using the Yates chi-square test (p = 0.0004 signiﬁcant ( = 3.676, p = 0.055). No normal controls were tested in

and 0.021 for the two patients, respectively). Both patients were these tasks, but one would expect normal observers to perform at

also impaired in naming digits, but they were disproportionately ceiling with all stimulus types, and thus it is unlikely that a similar

impaired in reading words and letters compared to single- and mul- difference in performance would be observed in normal individu-

tidigit numbers. Cohen and Dehaene (1995) also showed that their als in these exact tests. SP was also tested on a “perceptual speed”

patients’ accuracy in number reading depended on task demands. task where a single target was to be identiﬁed in a row of ﬁve

For instance, they noted signiﬁcantly more reading errors per digit stimuli from the same category. This task used letters, digits and

when patients were to read multidigit numbers as compared to ﬁgures as stimuli (in separate conditions). Performance was com-

single digits. This may be interpreted as an effect of task difﬁculty pared to two normal controls, and to a non-alexic patient with a

(reading two numbers at the same time is more difﬁcult than read- left parietal lesion. SP’s accuracy was similar to that of normal con-

ing one at a time). They also report that both patients were better trols for all three stimulus types, but he was slower to complete

at reading out digits in the context of a comparison task than in the test in all conditions. In addition, he was disproportionately

an addition task even though the exact same digits were used in slow in the letter condition, compared to the number and ﬁgure

both conditions. A similar effect was not observed when number conditions (Numbers 115s.; Letters 210s.; Figures 130s. The two

words (‘one’ ‘three’) were used as stimuli. It is important to note normal controls spent 60s. (SD = 0); 66s. (SD = 8.5); 62.5s. (SD = 3.5)).

that although performance with digits was better in the ‘reading No statistical analyses were presented in the paper, but using

for comparison’ than in the ‘reading for addition’ task, it was not Crawford and Howell’s (1998) test on the reported data (Perri et al.,

2288 R. Starrfelt, M. Behrmann / Neuropsychologia 49 (2011) 2283–2298

1996, Table 5) reveals that SP’s performance with letters was indeed 7 ms and 50 ms for normal observers). This enables the determi-

signiﬁcantly slower than that of the controls (p = 0.023). He was, nation of the perceptual threshold (the exposure duration below

however, signiﬁcantly slowed in the number (p = 0.01) and ﬁgure which recognition is zero), and processing speed (indicating the

(p = 0.018) conditions too. The control patient without alexia also speed of processing in items per second). In the whole report exper-

performed better with digits than with letters (both in accuracy and iment, ﬁve stimuli are presented simultaneously in either the left

RT), while his error rates in all three conditions were higher than for or right ﬁeld (usually left for pure alexic patients, as they commonly

SP. This makes the interpretation of the difference in performance have a right visual ﬁeld defect). Whole report also uses brief expo-

with letters and digits observed in SP difﬁcult, and, in the words of sures (30–200 ms in these case studies), and requires the subject

the authors: “This ﬁnding suggests that these types of tasks may to name as many of the presented letters/digits as possible. Again,

also be sensitive to deﬁcits other than the ones affecting reading several parameters can be measured using data from a single exper-

mechanisms” (Perri et al., 1996; p. 399). iment. Processing speed (now in the parafoveal visual ﬁeld) and

Ingles and Eskes (2008) present patient GM (case 89 in Table A1), visual apprehension span are the most important parameters in the

whom they classify as a letter-by-letter surface dyslexic (or Type current context. Using this method, Starrfelt et al. (2010) found that

2 pure alexic) patient. GM also had surface agraphia and anomia. NN’s visual apprehension span was uniformly reduced for letters

GM’s accuracy in naming digits was significantly superior to his and digits, but that his processing speed at fixation was signifi-

performance with letters in a RSVP (rapid serial visual presenta- cantly reduced only for letters (data are presented in Table 1). NN’s

tion) paradigm (p < 0.01). This pattern was also found in a group central processing speed was comparatively better with single dig-

of non-alexic patient controls, but the discrepancy between digits its than single letters, and the ratio between the two scores was

and letters was more pronounced in GM. In a speeded match- signiﬁcantly different from the pattern observed in the controls

ing task, GM’s RTs were elevated for letters compared with digits (assessed with Crawford and Howell’s test), strongly suggesting

(p < 0.01), a pattern not found in the control patients. Importantly, an asymmetry in letter/digit recognition. However, with respect

GM’s performance with digits in the matching task was impaired to overall accuracy in the single digit report task, NN was signiﬁ-

relative to patient controls, as was his performance with letters, cantly impaired compared to controls, indicating that his number

and his performance in both conditions is likely to be far below reading was also not entirely normal.

the normal level (no normal controls were tested in this study).

Again, while a difference in performance with digits and letters

2.3. A new analysis of published data

seems evident, the lack of a normal control group prevents us from

deciding whether the patient’s pattern of performance meets the

In a study of four pure alexic patients using the same TVA-based

criteria for a strong dissociation. Critically, as we discuss below,

experiments and analyses described above, Starrfelt et al. (2009)

the data presented might well be accounted for by an intrinsic

found impaired performance with both letters and digits for all

difference between letters and numbers (in terms of guess rate,

patients, compared to controls. Although no dissociations in sin-

visual confusability, nature of representation) rather than necessar-

gle item processing speed were discussed in this paper (and the

ily compelling an interpretation in terms of a letter/word-speciﬁc

patients are listed as showing no dissociation of performance in

module.

Table A1; cases 33–36), a closer inspection of the patients’ scores

Starrfelt et al. (2010) studied the performance of a pure alexic

reveals that three of the patients did show better performance with

patent (NN; case 90 in Table A1) on psychophysical tasks using

single digits than with letters in the single item report task. For

letters and digits as stimuli. Besides his alexia, NN showed no

the purposes of the current review, we have re-analysed the data

other impairments in language or cognition. This study was per-

from these three patients (cases 34–36 in Table A1), along with data

formed within the framework of the Theory of Visual Attention

from patient NN mentioned above (Starrfelt et al., 2010), to explore

(TVA; Bundesen, 1990), which is described in some detail here, as

possible dissociations using methods suggested by Crawford and

it is also relevant for the analysis presented in the next section.

Garthwaite (2005). This analysis ﬁrst compares the scores of the

TVA is a computational theory, and based on simple psychophysi-

patient to the control mean, using Crawford and Howell’s (1998)

cal tasks and mathematical modelling, several parameters in visual

test for a deﬁcit. As is evident from Table 1, and also as reported in

processing can be estimated. The two main parameters of interest

the original papers, all patients are signiﬁcantly impaired with let-

in the study of NN was visual processing speed (or recognition efﬁ-

ters, and all but NN are also impaired with digits. Then the patient’s

ciency) and visual apprehension span (the number of items that can

scores are converted to z-scores based on the control-group results,

be encoded into visual short term memory simultaneously), which

and the z-scores are compared using the Revised Standardized Dif-

were measured for letters and digits separately. NN participated in

ference Test (RSDT, Crawford & Garthwaite, 2005), and the criteria

two experiments (single item report and whole report), which both

for a dissociation (strong or classical) are met only if this differ-

make use of brief, masked stimulus presentation. After mask onset,

ence is statistically signiﬁcant. The patient and control data are

subjects have unlimited time to name the stimuli, and thus naming

presented in Table 1, and to illustrate the procedure, the analysis

latency does not affect scores within this paradigm. It should be

of patient NN’s results can be seen in Fig. 1.

noted that NN’s accuracy in naming letters and digits in free vision

The analysis reveals that none of the four patients fulﬁls the cri-

was perfect.

teria for either a strong or a classical dissociation in performance

In single item report, a letter or digit (in separate conditions)

between single letters and digits, because the difference between

is presented brieﬂy and then masked. Exposure durations are cho-

their standardized scores on the two tests is not statistically

sen individually, so that the subject’s performance from ﬂoor (zero

signiﬁcant. This means that a similar discrepancy in scores could be

correct) to ceiling (100% correct) can be assessed (usually between

found in the normal population with a probability of p > 0.05. Even

for the patient with the largest difference in performance (case 34

in Table A1; JH, Starrfelt et al., 2009), whose processing speed was

The standard deviation for controls was set to 0.1 in the number conditions, to

be able to perform this analysis. As we do not know the correlation between tasks in

the controls, we cannot use Crawford and Garthwaite’s (2005) methods for deciding In the single digit test, one normal control performed at a very high level, creating

whether a dissociation in present. Also, only two controls were tested, which further quite large standard deviations in the small control sample (N = 5). The authors thus

renders this analysis problematic. Thus, we do not know if SP’s performance with suggest that NN’s processing speed may well be reduced for numbers also, but that

letters was disproportionally impaired compared to his performance with digits. this failed to reach signiﬁcance due to the large SD’s in the control group.

R. Starrfelt, M. Behrmann / Neuropsychologia 49 (2011) 2283–2298 2289

Step 1: Compares performance on Task I (single letter processing speed) to controls

using Crawford & Howell’s (1998) test for a deficit.

NN’s score of 22 compared to 138 (SD = 48) for controls; t = - 2.35 (df = 4);

p(one-tailed ) = 0.03925

NN meets the criterion for a deficit on Task I.

Step 2: Compares performance on Task 2 (single digit processing speed) to controls

using Crawford & Howell’s (1998) test for a deficit..

NN’s score of 42 compared to 143 (SD = 47) for controls; t = - 1.826 (df = 4);

p(one tailed) = 0.07096

NN does not meet the criterion for a deficit on Task II.

Step 3: Compares the patient’s z-scores on Task I and II using the Revised Standardized

Difference Test (Crawford & Garthwaite, 2005), to test if the patient’s discrepancy

betwee n scores is greater than what could be expected in the control population.

Task I z-score = - 2.574

Task II z-score = - 2.00

t = -0.973 (df = 4); p(two-tailed) = 0.3858

The difference betwee n the patient’s standardized scores is not significant.

Step 4: Conclusion

The patient does not meet the criteria for either a strong or a classical dissociation.

Fig. 1. Illustration of procedure for analysing dissociations using Crawford & Garthwaite’s (2005) method. NN’s data compared to 5 controls (Starrfelt et al., 2010). Data are

listed in Table 1.

about 3.5 times slower for letters than digits, the difference does not that the observed difference in performance does not meet the

meet the statistical criteria for a dissociation: While she performs criteria for either a strong or a classical dissociation. Thus, there

signiﬁcantly differently from controls both with letters and digits, is no obvious evidence presented in any of the reviewed studies

the difference between her standardized scores on the two tasks that demands an explanation on the level of a cerebral area ded-

does not even approach signiﬁcance (z-digits = −2.5, z-letters = 4.1, icated to processing letters or words but not numbers. What we

t = 0.969, p = 0.36). The same is true for patient NN, whose process- are left with then, are some cases that show a trend dissociation,

ing speed for single digits was within the low normal range: As the which, as mentioned in the introduction, is generally thought of

difference between his performance on the two tasks is not signif- as a weak form of evidence for independence. Still, as the trend

icant (p = 0.39), the results do not fulﬁl Crawford and Garthwaite’s in all these studies points in the same direction (number read-

(2005) criteria for a strong or classical dissociation (see Fig. 1 for the ing is comparatively better than letter or word reading), one is

full analysis). In sum, these data show that even quite substantial left with the impression that number reading, in some cases, has

differences in performance with letters and digits may not consti- been spared to a degree, and that this asymmetry needs to be

tute a dissociation in the statistical sense, as such differences may explained.

also be found in the normal population. Indeed, one of our controls

had processing speed for letters at 83 letters/s and for digits at 123 3. Normal performance and visual similarity

digits/s, corresponding to z-scores of −2.25 and 0.261, respectively,

and a larger z-score difference than both of the patients mentioned 3.1. Normal recognition of letters and digits

above (NN and JH).

One possible explanation for the somewhat better reading of

2.4. Summary of results digits than letters in some pure alexic patients is that digits are

simply easier to process, and that this may be true even for nor-

The most important ﬁnding of our review is that there is not mal individuals. We start by investigating whether this is so and

a single well-documented classical dissociation between a pure report ﬁndings from a psychophysical study in which normal recog-

alexic patient’s reading of letters and digits in the literature to nition of single letters and single digits was recorded at varying,

date. “Normal”, “ﬂuent” or “unimpaired” reading of numbers have short exposure durations. We demonstrate that accuracy is higher

been reported, but only in papers where details of assessment for digits than letters and then, in the next section, we go on to

are not given, or where the tasks presented are not comparable investigate possible sources of this asymmetry.

in any meaningful way. While it remains a possibility that one

of these patients really does have normal number reading, this 3.1.1. Methods

would represent a departure from the general ﬁndings in this In order to explore the recognition performance of normal

review. Some papers do report strong dissociations between num- observers with letters and digits, we have analysed data from a sin-

ber and letter reading, but when stringent statistical criteria for gle item report task that was part of a larger study of visual attention

dissociations are applied to the data from these studies, we ﬁnd (unpublished data, the study was conducted by the ﬁrst author in

2290 R. Starrfelt, M. Behrmann / Neuropsychologia 49 (2011) 2283–2298

Table 2

A comparison of mean (SD) proportion correct responses in the single item report task for digits vs. letters for the six exposure durations, and the overall mean correct. Data

are from 20 normal subjects, and were analysed using paired-samples t-test.

Digits Letters t (df = 19) p

13 ms 0.39 (0.25) 0.29 (0.22) 4.3 <.001

20 ms 0.74 (0.19) 0.58 (0.26) 5.6 <.001

27 ms 0.91 (0.09) 0.78 (0.21) 3.5 <.01

33 ms 0.95 (0.05) 0.88 (0.14) 2.8 <.02

40 ms 0.97 (0.03) 0.91 (0.11) 3.0 <.01

53 ms 0.98 (0.03) 0.97 (0.05) 0.77 .454

Total 0.82 (0.10) 0.73 (0.15) 4.8 <.001

collaboration with T. Habekost and K.I. Karstoft at the Center for differs for the two types of symbols. With pure guessing, the chance

Visual Cognition, Copenhagen University). The subjects (university of getting the correct digit is 1/10, while it is only 1/26 in letters,

students, N = 20, 10 female) performed a single item report task thereby affording better guessing with digits than letters. Guess-

similar to the one described in section 2.2 (Starrfelt et al., 2009, ing may be invoked under difﬁcult encoding conditions, such as

2010). Digits (0–9) and upper case letters (A–J) were run in sepa- the very brief exposure conditions reported in section 3.1. How-

rate blocks of 120 trials in a ABBA–BAAB design (N = 480 trials per ever, we do not usually proceed with wild (unconstrained) guessing

stimulus type per subject; half the subjects did letters ﬁrst, half when identifying visual stimuli. Rather, we commonly have at least

did digits ﬁrst). Before the start of the experiment, the participants some information available, on which we base a ‘qualiﬁed percep-

were shown the 10 digits and the 10 letters, and were asked to read tual guess’, and the degree of within-category visual similarity is

through them. Subjects were told that only the letters A–J would be likely to be important in this process. Speciﬁcally, under condi-

used as stimuli in the letter condition. Stimuli were computer gen- tions of insufﬁcient evidence for a particular target, the number of

erated, and did not conform to any known typefont (but resemble within-category competitors (other letters or digits) that resemble

the Rumelhart and Siple font, see Starrfelt et al., 2010 for images the target become important. Here we explore whether visual sim-

of stimuli and mask). Stimuli were presented in white on a black ilarity or discriminability differs between letters and digits, using

background. two different methods: First, we compare image-based pairwise

On each trial, a single item, either digit or letter, was ﬂashed similarity for the two categories and, second, we measure the num-

brieﬂy using six different exposure durations randomly inter- ber of within-category competitors as a function of different levels

mixed (range 13–53 ms) and immediately followed by a pattern of discriminability. In the latter analysis, we also take into consider-

mask for 500 ms (screen refresh rate was set so there was no ation that, as a raw number, there are potentially more competitors

delay between stimulus and mask). Subjects were to report the for letters than for digits.

stimulus if they were fairly certain of its identity and there was

no requirement to report the stimulus at speed. Accuracy was 3.2.1. Methods

recorded.

There are many ways to compare the image properties of let-

ters and digits – for example, one might count absolute strokes

per symbol or the number of curved vs. straight strokes. To avoid

3.1.2. Results

making any a priori assumptions on how to carve up the symbols,

Mean accuracy scores were computed for letters and digits

we computed letter and digit image-based similarity using one of

separately for all six exposure durations, and an overall correct

the simplest methods for template matching and similarity esti-

score (summed over all exposure durations) was also computed for

mation: normalized cross-correlation (Lewis, 1995). To do so, we

letters and digits separately. Data are presented in Table 2. A com-

cross-correlated each pair of symbol images, and found the peak

parison using a paired-samples t-test shows that overall accuracy

of the cross-correlation map, that is, the horizontal/vertical trans-

was signiﬁcantly higher for digits than letters (t19 = 4.78, p < .001).

lation of one image with respect to the other that maximizes their

Also, subjects performed signiﬁcantly better with digits than letters

alignment. We then recorded the respective correlation coefﬁcient

on average in the ﬁve conditions with the shortest exposure dura-

(i.e., degree of overlap between the two images, see Fig. 2a) as an

tions (13–40 ms; see Table 2). In the condition with the longest

estimate of similarity: the higher the correlation, the more similar

exposure duration (53 ms), there was no difference in accuracy for

the images of the two symbols. This provides an estimate of max-

letters and digits, as performance with both was at ceiling, with

imal similarity between two symbols, and the cross-correlation

mean accuracy of 97% and 98%, respectively.

coefﬁcient can be thought of as the worst-case scenario when dis-

In sum, these ﬁndings indicate that digits are signiﬁcantly easier

crimination of a symbol from its competitors is most difﬁcult. This

to identify than letters for normal subjects under conditions of brief,

analysis was done separately for letters and digits displayed in high

masked exposure, that is, when the perceptual information about

contrast (black against white background) and it was done for two

the stimuli is limited. Also, the ﬁndings show that normal observers

reach ceiling performance with both letters and digits at relatively

brief exposure durations. This pattern of results is important as it

suggests that the tendency towards better performance with digits

than letters in pure alexic patients might reﬂect an ampliﬁcation of

a normal difference in symbol processing. This still, however, begs

the question of where this difference originates.

Fig. 2. Illustration of physical overlap (cross-correlation) between symbols. On the

3.2. Visual similarity for letters and digits

left we have overlaid an Arial font ‘9’ (dark grey) on ‘0’, and ‘0’ (dark grey) on ‘9’;

these symbols have a very high cross-correlation of 0.85. On the right an illustration

There are a number of potential sources for the asymmetry in

of overlaying ‘v’ (dark grey) with ‘n’, and ‘n’ (dark grey) on ‘v’, respectively; these

performance with letters and digits. One is that the guessing rate two letters have a quite low cross-correlation of 0.37.

R. Starrfelt, M. Behrmann / Neuropsychologia 49 (2011) 2283–2298 2291

Fig. 3. Illustration of the number of competitors for letters and digits in Arial (left panel) and Times (right panel) fonts, based on the pairwise cross-correlation analysis of

the symbols. ‘Mean correlation/threshold’ refers to the interval of pairwise cross-correlation coefﬁcients, and within each interval the average number of competitors per

symbol are shown, separately for letters and digits. The higher the number of competitors per symbol, the higher the confusability.

Table 3

different fonts, Times and Arial. Letters were in lower case as this is

Average cumulative frequency of letters and digits (per symbol) in the different

the more common case encountered in text reading and captures

ranges of cross-correlation for (a) Times and (b) Arial. See also Fig. 2 for a graphical

the perceiver’s task more veridically. illustration.

Cross-correlation threshold Letter Digit

3.2.2. Results

(a)

In Arial, the mean similarity (cross-correlation) across symbol

0.9 0 0

pairs was 0.58 (SD = 0.15) for letters and 0.61 (SD = 0.15) for digits,

0.8 0.5 0

and the corresponding values for Times were 0.57 (SD = 0.12) and

0.7 4 0.2

0.55 (SD = 0.08). A t-test conducted within each font reveals no sig- 0.6 8.9 2.4

0.5 16.5 6.4

niﬁcant differences between the two symbols types, indicating that,

0.4 23.4 9

with this measure, the within-category similarity for letters and

0.3 24.9 9

digits is about equal. This result does not support an explanation

0.2 25 9

of the behavioural asymmetry based on pairwise symbol similarity 0.1 25 9

(although it is possible such a difference might exist for types of 0 25 9

font other than those tested here). However, this comparison may (b)

not be the most informative, since letters and digits are normally 0.9 0.2 0

0.8 2.2 1.2

recognized as part of the entire set of letters and digits, respectively,

0.7 6.2 3.2

rather than as a forced choice between just two alternatives. Thus,

0.6 9.6 4

a more meaningful comparison should take into account the visual

0.5 15.4 6

similarity of all potential competitors and should also factor in the 0.4 22.7 9

category set. 0.3 24.9 9

0.2 25 9

Therefore, the procedure we adopted was to determine the

0.1 25 9

average number of competitors as a function of discriminability

0 25 9

threshold separately for letters and digits, where discriminability is

indexed by a given cross-correlation threshold. For example, using

the pairwise similarity values previously computed, we ﬁnd that of the items in the total stimulus set. For instance, if we index

the letter ‘a’ has two competitors if the discriminability threshold confusability as the number of symbol pairs achieving a similar-

lies in the 0.7–1 range, i.e., two letters have a pairwise cross- ity score above a 0.7 threshold, then the confusability of letters

correlation with ‘a’ higher than 0.7 (‘n’ and ‘u’). This rises with eight is about twice the confusability for digits. In conclusion, when an

competitors if the discrimination threshold is lowered to 0.6–1 observer encounters a single letter, there are more possible com-

(including ‘o’, ‘c’ and ‘e’), another 12 for 0.5–1 (such as ‘m’, ‘p’ and petitors (average of all pairwise confusability measures) than is

‘b’), and all letters when we drop the discrimination threshold to 0.4 the case for digits and this is true even when one holds constant

(adding the last two ‘y’ and ‘v’ to the set). For each symbol, we added the discriminability threshold to equate for the set size for let-

up the average number of confusable stimuli within each thresh- ters and digits. On this basis, single symbol identiﬁcation should

old interval, as plotted in Fig. 3 for the Times and Arial symbols (the be more difﬁcult for letters than for digits (at least for the fonts

actual numbers are shown in Table 3). Looking at the Times ﬁgure, we have investigated here), when the amount of available visual

we see that, on average, every letter starts with no competitors if information is limited.

the discriminability threshold is high, but acquires systematically

more competitors as we lower the discriminability threshold (e.g., 4. Discussion

4 within the 0.7–1 range) and ends up with the entire letter set

when the threshold is low enough. The same is true for the Arial This aim of this paper is to examine whether number reading

font analysis. can be intact in pure alexia. If so, this would support the hypoth-

The important point here is that letters have more competitors esis that pure alexia is a domain-speciﬁc deﬁcit in visual word-

than digits for any given discriminability threshold below the most and letter-processing. Understanding the mechanism(s) underly-

optimal. This has to be the case with low thresholds, because there ing pure alexia would not only elucidate the origin of this particular

are more letters than digits and thus many more possible com- disorder but, more generally, would shed light on an ongoing

petitors. However, Fig. 3 shows that this asymmetry is also present debate about the nature of brain-behaviour organization. Specif-

when discriminability threshold is high enough to exclude many ically, much recent literature argues in favour of a circumscribed

2292 R. Starrfelt, M. Behrmann / Neuropsychologia 49 (2011) 2283–2298

word recognition system (the so-called Visual Word Form Area; reading of numbers and even of number words, with relatively

VWFA), which is specialized for or even selectively dedicated to preserved word reading (e.g., Marangolo, Nasti, & Zorzi, 2004;

processing alphabetical input (Cohen et al., 2003, 2004). Intact see Piras & Marangolo, 2009 for an overview). This has led to a

number reading in patients with pure alexia, whose damage is discussion about a separate semantic system for numbers, and

usually to the VWFA and surrounding structures (Leff et al., 2006) even a separate output lexicon for number names (e.g., Marangolo

would provide compelling evidence for cerebral specialization for et al., 2004), and to theorizing about the relation between language

visual word recognition. Such a ﬁnding would be particularly infor- and number processing at levels involving conceptual (rather than

mative because there are so many similarities between letters and perceptual) representations (e.g., Gelman & Butterworth, 2005).

digits; hence if the deﬁcit affected letters selectively, this would However, to our knowledge, there are no reports of a number

strongly endorse the view of a domain-speciﬁc orthographic sys- reading deﬁcit with spared letter reading following damage to

tem. An alternative proposal is that the mechanism mediating perceptual/temporo-occipital areas of the brain, akin to the locus

visual word processing is somewhat more general-purpose and underlying pure alexia. Thus, there is no extant evidence from

capable of representing multiple types of input, but is perhaps lesion studies, including the ones we have reviewed, for a strong or

optimized for word recognition (e.g., Behrmann, Nelson et al., classical single dissociation (and therefore no double dissociation)

1998; Behrmann, Plaut, & Nelson, 1998; Starrfelt & Gerlach, 2007; between visual identiﬁcation of letters and numbers. And yet there

Starrfelt et al., 2009). On one recent account, the VWFA serves is this trend. . .

to bridge between higher-order visual areas and language areas One possible account of a difference in performance between

and, while this cortical region is not dedicated for letter/word any two tasks, and indeed the ﬁrst explanation to consider for

recognition per se, the bidirectional connectivity between these a single dissociation observed in brain injured subjects (Shallice,

regions fine-tunes the VWFA for maximal efficiency in representing 1988), is that one task is merely more difficult than the other. This

alphanumeric symbols (Plaut & Behrmann, in press). might be the case here: When we investigated the performance

To assess the evidence for a dissociation between number and of normal observers required to identify letters and digits under

letter reading in pure alexia, we undertook a comprehensive review brief exposure durations, we found better identiﬁcation of digits

of the literature on pure alexia from 1892 to 2010. We adopted a than letters at all but the longest exposure duration, at which point

stringent classiﬁcation system based on Shallice’s (1988) deﬁni- performance was approaching ceiling in both cases. Our interpre-

tions of dissociations, coupled with recent guidelines for assessing tation of this observed superiority for digits in normal subjects is

the statistical legitimacy of the empirical observations (i.e., testing that the relative preservation of digit over letter processing in pure

statistically if the observed difference in performance is larger than alexic patients might simply reﬂect damage to a common system

what could be expected in the normal population, Crawford et al., for processing of alphanumeric symbols. The difference in perfor-

2003). Interestingly, we did not ﬁnd a single study documenting a mance observed in patients seems larger than that demonstrated in

classical dissociation in which letter or word reading is impaired normal subjects, perhaps because the difference observed in nor-

while number reading is fully preserved. Several studies do report mals may be be exacerbated following brain injury; thus, when the

better performance with digits than letters (40 of the 76 reviewed system is perturbed, the more vulnerable of the two domains is dis-

studies, reporting a total of 44 patients), but most of these (30 proportionately affected. This still leaves the question, however,

studies reporting 33 patients) do not report sufﬁcient details about of why digits are more easily identiﬁed than letters when visual

methods or results for us to evaluate the status of the reported dis- information is limited.

sociations. Some studies show a statistically signiﬁcant difference One contributing factor could be the difference in guessing rate,

between processing of letters and numbers in pure alexic patients given that there are simply more letters than digits. When visual

(e.g., Cohen & Dehaene, 1995; Ingles & Eskes, 2008), but none of information is limited or degraded, and thus accuracy < 100%, the

these studies conform to stringent statistical criteria for a strong probability of correct guessing may inﬂuence the proportion of cor-

dissociation (Crawford et al., 2003). In some cases (N = 10), letter rect responses. In our study reported in section 3.1, the normal

and number reading are both reported to be intact, but, in nine of participants were informed that the target digit would be one of

these, a control group is lacking. The exception is a study by Rosazza 10 digits and the target letter would be one of 10 possible letters.

et al. (2007). They show that their patient FC had normal reaction They were familiarized with the stimuli prior the experiment to

times in naming both letters and digits, but do not present accuracy minimize the difference in guessing rate between the two stimu-

data for these tasks. In other tests of visual recognition, the patient lus types. Even under these conditions, performance was better for

was impaired. digits than letters. There is the possibility that the entire character

Thus, within the bounds of our stringent framework, we are left set, rather than the speciﬁc subset, might automatically be acti-

with a few studies (10 studies reporting 11 patients) that show vated or accessible even under these ‘equated’ conditions, a point

a trend dissociation, with number reading being less affected than that illustrates how difﬁcult it is to create comparable experimen-

letter and word reading. This does not constitute evidence for selec- tal conditions for letters and digits. There is also the possibility

tivity or cerebral specialization for visual word processing, but it is that digits are simply easier to recognize than letters regardless

remarkable that the trend points in the same direction in almost of the size of the character set, but perhaps because of structural

all cases of pure alexia; performance with digits is better than with aspects. On the basis of our analysis of image statistics, where we

letters. performed a normalized cross-correlation pairwise for all letters

Inferences about separability of systems not only demand a and for all digits, we suggest that the superiority of digits over let-

single dissociation (independent of its strength) but also the com- ters may be related to the number of visually similar competitors for

plementary pattern of preservation and loss. In some patients with the two sets of symbols. This analysis showed that digits are more

more general language disorders following lesions to temporo-

parietal areas in the left hemisphere, a pattern of performance

opposite to the trend in pure alexia, has been reported: impaired

Searching the literature for a selective number reading deﬁcit is more difﬁcult

than a search for pure alexia, as there is no syndrome label for a primary deﬁcit

in number reading with intact writing and calculation, or for a pattern of perfor-

Judd et al. (1983) report a rare exception: They note that in their patient “num- mance with impaired number reading and intact letter reading. We have performed

ber reading was worse than his letter reading” (p. 445), but no further details are a search for the keywords ‘number alexia’ and ‘number agnosia’ in Web of Science,

provided. which retrieved no references.

R. Starrfelt, M. Behrmann / Neuropsychologia 49 (2011) 2283–2298 2293

discriminable than letters in the sense that letters have more visu- remains ill-speciﬁed. One possibility is that the lesion impacts ﬁne-

ally similar competitors even when the discriminability threshold grained discriminations between homogenous symbols. The siting

is very high, e.g., when only symbols cross-correlating 0.7 or more of the lesion associated with pure alexia in what would be the ante-

with the target symbol are deﬁned as competitors. This index of rior extrapolation of the fovea (Hasson, Levy, Behrmann, Hendler,

discriminability would be expected to be most important for sym- & Malach, 2002; Plaut & Behrmann, in press) is consistent with

bol identification when the amount of available visual information this. Suffice it to say that this difficulty in pattern recognition can

is suboptimal, as is the case when stimulus presentation is experi- slow down the speed of visual processing, reduce visual span, and

mentally degraded, or – as may be the case in pure alexia – when give rise to visual confusions and various crowding effects. This

it is degraded by a lesion affecting visual perception. Thus, even if problem in pattern discrimination functionally compromises the

both numbers and letters were processed by a single system for representations of both letters/words and numbers, but the emer-

visual recognition, and this system was affected by the lesion caus- gent pattern of relatively less impaired digit over alphabetic symbol

ing pure alexia, the differential demands of processing numbers vs. recognition may nevertheless persist.

letters could give rise to the apparent superiority of numbers over This paper focuses on dissociations between digits and letters,

letters. but it would be fruitful to explore the extent to which the sug-

We have focused thus far on perceptual differences between let- gested perceptual deﬁcit in pure alexia affects processing of other

ters and digits, but there are also high level, semantic differences classes of visual stimuli. For instance, the ability of pure alexic

between numbers and letters that may affect perceptual process- patients to read musical notation – which we would expect to be

ing of these stimuli. Digits have a concrete referent (amount), and affected by the perceptual deﬁcit suggested above –islargely unex-

thus may be associated with semantic value or semantic repre- plored. The few studies on record investigating this issue present

sentations in a different way than single letters, that are commonly contradictory ﬁndings (see Horikoshi et al., 1997 for an overview),

only meaningful in strings (see Cohen & Dehaene, 1995). This could much in the same way that the literature on number reading has

give rise to greater top-down effects on the speed and accuracy seemed contradictory. Picture naming has been quite extensively

of digit recognition compared with letter recognition, particularly studied in pure alexia, but here too the ﬁndings are inconsistent.

in cases where visual information is limited. In a similar vein, it Normal picture naming has been reported in some pure alexic

has been suggested that top-down effects on word recognition in patients with regards to accuracy (e.g., Gaillard et al., 2006; Maher

pure alexia, both in reading and lexical decision tasks, may inter- et al., 1998; Starrfelt et al., 2010), while in tasks measuring RTs

act with the severity of the patients’ reading problems (Behrmann, (e.g., Starrfelt et al., 2009) or where subtle object discrimination is

Plaut et al., 1998; Roberts, Lambon Ralph, & Woollams, 2010). The required (Sekuler & Behrmann, 1996; Starrfelt et al., 2010), perfor-

general hypothesis is that for mildly affected patients, bottom-up mance is commonly impaired. As the severity of visuo-perceptual

activation is sufﬁcient to generate the correct response, while for problems in pure alexia has been found to be related to reading

moderately affected patients in whom bottom-up input is slow skills (Mycroft, Behrmann, & Kay, 2009), further exploration of the

and error prone, top-down effects may signiﬁcantly aid recogni- nature of these problems seems important. In general, investigat-

tion. For the most severely affected patients, because the bottom-up ing apparent dissociations, e.g., between recognition of words and

activation generated by the stimuli will be too weak, top-down pro- other objects in pure alexia, using sensitive tests and the statis-

cessing will be of no assistance. It is possible that such an account tical methods applied here, could provide important insight into

might also explain the trend towards better reading of digits, which the workings of the cerebral systems underlying visual recogni-

may have more semantic (top-down) support, than letters. This tion and offer further constraints on theories of brain-behaviour

account makes a counterintuitive prediction that there will be rel- correspondences.

atively greater divergence between digit and letter recognition in

more moderately than more mildly affected pure alexic patients

Acknowledgements

since the former will beneﬁt more from top-down support for

digits. Thus, the relationship between performance and the sug-

This research was supported by a grant from the Center of Excel-

gested visuoperceptual impairment in pure alexia may not always

lence Program at the University of Copenhagen to RS/Center for

be straightforward, and we suggest that the differences between

Visual Cognition, and a grant from the National Science Founda-

letter and digit processing may be explained by an account in which

tion to MB. Thanks to Christian Gerlach and Thomas Habekost for

stimulus characteristics (e.g., visual discriminability and semantic

comments on earlier versions of this manuscript, and to Fakutsi

value) interact with task type (e.g., naming vs. magnitude compar-

for valuable insights. Thanks to Adrian Nestor for computing the

ison) and impairment severity.

cross-correlation data.

Other explanations also remain possible. For instance, the

hypothesis that number reading is more bilaterally distributed

than letter and word reading, and therefore more likely to be rel- Appendix A.

atively preserved in pure alexia (Cohen & Dehaene, 1995, 2000)

cannot be ruled out by the presented ﬁndings. However, even if A complete list of references mentioning alexic patients’ per-

there is a right hemisphere system with capabilities for number formance with numbers are listed in Table A1. Studies are listed

recognition, it seems clear from the present review that the left in groups, deﬁned by whether a dissociation in performance with

hemisphere damage associated with impaired letter processing in letters and digits seems to be present, and how well supported

pure alexia also affects number reading. We suggest that this is this dissocation is by the presented data. The column labels and

caused by a visuoperceptual impairment that impacts alphanu- abbreviations used are listed and explained below the table.

meric processing more generally. Determining the nature of this Aetiology: The reported cause of the observed reading deﬁcit.

visuoperceptual impairment has been the subject of many pre- Abbreviations: (h) = hemorrhagic/hematoma; (surg) = following

vious papers (for some examples, see Behrmann, Nelson et al., surgery; AVM = Arterio-venous malformation; CJD = Creutzfeldt-

1998; Farah & Wallace, 1991; Fiset, Gosselin, Blais, & Arguin, 2006; Jakob disease; HSV = Herpes simplex virus: MS = Multiple sclerosis;

Starrfelt et al., 2010). While there seems to be general consensus NR = not reported; TBI = Traumatic brain injury; ? = uncertain

that the problem arises at higher levels of visual representation diagnosis.

rather than in early visual cortices, and the typical lesion site is Diagnosis: The label for the observed reading disorder pro-

consistent with this claim, the exact nature of the impairment vided by the authors of the original papers. Abbreviations:

2294 R. Starrfelt, M. Behrmann / Neuropsychologia 49 (2011) 2283–2298

Table A1

Author (year; patient) Aetiology Diagnosis Lesion Visual ﬁeld Agraphia Aphasia

Ia. Number and letter reading both impaired at roughly same level.

1 Joy (1947; CS) Trauma AWA(ag) NR URQ no an

2 Kinsbourne and Warrington (1963; h surg SD L STG; ITG RH (ambl.) no no

Mrs.B)

3 Collignon (1972; Mrs.B.L.) Stroke PA L posterior RH no NR

4 Caplan and Hedley-Whyte (1974; Stroke AWA L temp-occ; RH slight an

XX) Scc; Hi; Th; LG

5 Woods and Pöppel (1974; XX) TBI or stroke AWA NR RH (partial) no nam

6 Vincent, Sadowsky, Saunders, and Meningioma AWA L inf temp-occ. None no no

Reeves (1977; XX)

7 Holtzman, Rudel, and Goldensohn Meningioma(surg) AWA L inf temp-occ; None. no no

(1978; XX) AnG

8 Levin and Rose (1979; XX) Meningioma(surg) AWA L occ-temp; Scc RH slight an

9 Shipkin, Grey, Daroff, and Glaser Atrophy AWA Diffuse LH no no

(1981; XX)

10 Judd, Gardner, and Geschwind Stroke(h) AWA L occ-temp RH yes nam; comp

(1983; BL)

11 Henderson, Friedman, Teng, and Stroke(h) PA L FuG; ITG; O2 None no no

Weiner (1985; XX)

12 Henderson (1987; Pt.1) NR PA NR NR NR NR

13 Henderson (1987; Pt.2) NR PA NR NR NR NR

14 Henderson (1987; Pt.3) NR PA NR NR NR NR

15 Coslett and Saffran (1989; JG) Multiple PA LG; or; Scc RH no no

strokes

16 Coslett and Saffran (1989; TL) Stroke PA L occ; fmj RH NR no

17 Coslett and Saffran (1989; JC) Stroke PA L occ.; fmj; pic RH yes no

18 Farah and Wallace (1991; TU) AVM/stroke PA L temp RH NR an

19 Iragui and Kritchevsky (1991; XX) Stroke AWA L subangular None yes nam; comp

20 Price and Humphreys (1992; EW) Stroke LBL NR RH NR dys

21 Price and Humphreys (1992; HT) TBI(surg) LBL L par-occ None. yes nam

22 Coslett, Saffran, Greenbaum, and Stroke PA L med occ; RH NR nam

Schwartz (1993; JWC) PHG; pic

23 Buxbaum and Coslett (1996; JH) Stroke LBL(deep) L inf temp-occ; RH no no

PHG; LG

24 Sekuler and Behrmann (1996; MA) TBI PA Bilat RH surface nam(RT)

25 Sekuler and Behrmann (1996; DS) Stroke PA L occ URQ no nam

26 Sekuler and Behrmann (1996; MW) Stroke PA L occ ? NR nam(RT)

26 Behrmann, Plaut et al., (1998; EL) Stroke PA L inf temp; lat URQ no nam(RT)

temp; dor par

27 Miozzo and Caramazza (1998; GV) Stroke PA L occ-temp; RH no Optic aph.

Scc; R par

28 Cohen and Dehane (2000; VOL) Stroke PA L inf med RH no nam; comp

occ-temp

29 Dalmas and Dansilio (2000; AA) Stroke AWA(vg) L ccs; LgG; RH no no

FuG; Scc; Hi;

PHG; Th

30 Lambon Ralph et al. (2004; FD) Stroke(h) PA Multiple: L occ; RH yes an

L par-temp; R

par; R occ

31 Armbruster and Wijdiks (2006; XX) Stroke PA L occ-temp RH no nam

32 Rosazza et al. (2007; LDS) Stroke PA L ITG; FuG; ots; RH no no

O4 + bilat front

33 Starrfelt et al. (2009; BA) TBI(h) PA L ITG; FuG; RH no nam(RT)

PHG

34 Starrfelt et al. (2009; JH) AVM(h) PA L 17; O2; O3; RH no nam(RT)

O4; LgG; MTG;

ITG; FuG

35 Starrfelt et al. (2009; JT) Stroke PA L 17; O2; O3; RH no nam(RT)

O4; LgG; ITG;

FuG; PHG

36 Starrfelt et al. (2009; TJ) Stroke PA L O2; FuG; PHG none no nam(RT)

Ib. Number and letter reading both reported normal or intact.

37 Greeblatt (1976; XX) AVM(surg) AWA(trans) L temp-par None (R no no extinction)

R. Starrfelt, M. Behrmann / Neuropsychologia 49 (2011) 2283–2298 2295

Table A1 (Continued)

Author (year; patient) Aetiology Diagnosis Lesion Visual ﬁeld Agraphia Aphasia

38 Rosati, Debastiani, Aiello, and Stroke(h) AWA L ITG; O2 URQ no no

Agnetti (1984; XX)

39 Caffarra (1987; MR) Stroke(h) AWA L post LOTG None no no

40 Daniel, Bolter, and Long (1992; KV) Stroke(h) AWA L temp-par-occ URQ + yes an LoRQ

(partial)

41 Warrington and Langdon (1994, Stroke SD L occ-par L eye nasQ no no

2002; ROC)

42 Dogulu, Kansu, and Karabudak MS AWA L occ.; Scc; RH no no

(1996; XX) Bilat ic

43 Mao-Draayer and Panitch (2004; MS AWA L occ; Scc RH no no

XX)

44 Verma, Singh, and Misra (2004; XX) Neurocysticercosis AWA(trans) L occ RH no no

45 Celebisoy, Sagduyu, and Atac Stroke(h) PA L occ RH no no

(2005; XX)

46 Rosazza et al. (2007; FC) Stroke PA L OcP; LgG; RH no no

PHG; O4; R O2

+ O4

IIa. Possible dissociation (numbers > letters). Tasks not directly comparable.

47 Warrington and Shallice (1980; Stroke(h) WFD L temp-par RH no an

RAV)

48 Patterson and Kay (1982; TP) Angioma(h) LBL L occ-temp-par RH surface an

49 Leegaard, Riis, and Andersen Stroke(h) PA L dors lat occ URQ no wfd

(1988; XX)

50 Di Pace, Guariglia, Judica, Spinelli, TBI LBL Bilat front; L R scotoma NR no

and Zoccolotti (1995; GM) occ; ox; R occ right eye;

left eye

10/1200

51 Chialant and Caramazza (1998; MJ) Stroke LBL Multiple wm RH no no

lesions; or; cc

IIb. Possible dissociation (numbers > letters). Poorly described or tested.

52 Geschwind and Fusillo (1966) Stroke PA/AWA L ccs; Scc; or; RH no no

Hi; VPL

53 Lee (1966) Trauma? AWA NR RH no an

54 Cumming, Hurwitz, and Perl Stroke AWA L occ; Scc RH no no

(1970; WL)

55 Benson, Brown, and Tomlinson Stroke AWA L post med RH no an

(1971; Case 1)

56 Goldstein, Joynt, and Goldblatt CO- AWA Diffuse None no no

(1971; XX) poisoning

57 Wechsler, Weinstein, and Antin Stroke(h) PA L occ-temp RH no an

(1972; Case 1)

58 Wechsler et al. (1972; Case 2) AVM(h) PA L post RH no nam

temp-occ

59 Wechsler et al. (1972; Case 3) Stroke(h) PA L occ-par RH no an

60 Greenblatt (1973; XX) Glioblastoma AWA L FuG; LgG; wm None yes nam

61 Fincham, Nibbelink, and Metastases AWA L PrG; SMG; R LH no no

Aschenbrener (1975; XX) multiple

62 Luhdorf and Paulson (1977; XX) NR PA NR Incomplete no nam

63 Karanth (1981; NR) NR PA NR RH slight nam

64 Winkelman and Glasson (1984; Stroke AWA R temp; occ; LH no nam/wfd

XX) par; Th; fmj;

Scc;

65 Lang (1985; XX) Stroke AWA L MOTG; LOTG; URQ slight nam/wfd

(multiple) O2; PHG; Scc;

basal ganglia

66 Pena Casanova, Roig Rovira, Stroke AWA L med temp LH no no

Bermudez, and Tolosa Sarro (1985;

AR)

67 Regard, Landis, and Hess (1985; Metastases PA L occ; front; R RH no nam

XX) par

68 Marks and DeVito (1987; Case1) Stroke AWA L inf occ; post None no wfd

temp; R

occ-temp

69 Pillon, Bakchine, and Lhermitte Stroke AWA R occ-temp LH no nam

(1987; XX)

70 Katz (1990; Case 1) Stroke LBL Multiple RH no an

71 Katz (1990; Case 2) Stroke LBL Multiple RH yes an

72 Dufﬁeld, Desilva, and Grant (1994; Stroke AWA L occ None no no XX)

2296 R. Starrfelt, M. Behrmann / Neuropsychologia 49 (2011) 2283–2298

Table A1 (Continued)

Author (year; patient) Aetiology Diagnosis Lesion Visual ﬁeld Agraphia Aphasia

73 Erdem and Kansu (1995; XX) HSV AWA L temp (OTG); L None no an; wfd

front-temp

74 Lanzinger, Weder, Oettli, and Fretz Stroke GA/SD L inf med temp URQ no an

(1999; XX)

75 Goodglass, Lindﬁeld, and Stroke PWB L RH no an

Alexander (2000; RH) occ-temp-par;

Scc; L FrP

76 Imtiaz, Nirodi, and Khaleeli (2001; Stroke AWA L par-occ URQ no nam; wfd

XX)

77 Leff et al. (2001; AR) Stroke(h) PA L med lat occ None no no

78 Adair, Cooke, and Jankovic (2007; CJD PA L temp-par None no nam; comp

XX) (atrophy)

79 Miglis and Levine (2010; XX) Stroke(h) PA L post inf temp None no nam

III. Trend dissociation (numbers > letters).

80 Dejerine (1892; MC) Stroke AWA L occ; Scc URQ, no no

LoRQachrom

81 Ajax, Schenkenberg, and Stroke? PA L O4; LgG; RH no no

Kosteljanetz (1977; Mr A) PHG; Hi; Scc;

fmj

82 Landis et al. (1980; XX) Glioblastoma AWA L temp-par-occ RH no no

83 Grossi et al. (1984; XX) Stroke AWA L occ; RH NR no

temp-occ; Scc;

84 Maher et al (1998; VT) Stroke PA L temp; LgG; RH no no

FuG; Cun; Cb

85 Larsen et al. (2004; EA) Stroke GA L temp-occ; Hi; RH slight an; wfd

Scc

IV. Strong dissociation (numbers > letters).

86 Cohen and Dehane (1995; GOD) Stroke PA L ccs; LgG; FuG RH no wfd

87 Cohen and Dehane (1995; SMA) Stroke PA L ccs; LgG; RH no no

FuG; Th

88 Perri et al. (1996; SP) Stroke(h) LBL L occ-par RH slight nam

89 Ingles and Eskes (2008; GL) Stroke(h) LBL(sur) L temp-occ Remitted surface an

URQ

90 Starrfelt et al. (2010; NN) Stroke(h) PA L 17; O2; inf URQ no no

LgG; FuG; SFG;

(ag) = agnosic; (deep) = deep dyslexia; (surf) = surface dyslexia; radiation; ots = occipito-temporal sulcus; ox = optic chiasm;

(trans) = transitory; (vg) = visuographemic; AWA = alexia with- PHG = parahippocampal gyrus; pic = posterior limb of inter-

out agraphia: GA; Global alexia; LBL = letter by letter read- nal capsule; PrG = precentral gyrus; Scc = splenium of corpus

ing; PA = pure alexia; PWB = pure word blindness; SD = spelling callosum; SFG = superior frontal gyrus; SMG = supramarginal

dyslexia; WFD = word form dyslexia. gyrus; STG = superior temporal gyrus; Th = thalamus;

Lesion: Lists the reported anatomical location of cerebral VPL = ventroposterolateral nucleus of the thalamus; 17 = striate

lesions. Some patients have multiple lesions. The methods used to area/primary visual cortex.

determine lesions vary from post mortem examinations, through Visual ﬁeld: Lists the type of visual ﬁeld defect reported in

angiograms and computerized tomography, to magnetic resonance the original papers. Note that method of assesment of visual

imaging, and the level of detail in lesion description varies accord- ﬁelds differs between studies, and some patients reported to

ingly. Abbreviations for speciﬁc areas are based on Mai et al. (1997). have “none” may have had defects not detected by e.g., con-

General abbreviations: L = left; R = right; bilat = bilateral; frontation testing. Abbreviations: R = right; L = left; U = upper;

inf = inferior; sup = superior; post = posterior; med = medial; Lo = lower; achrom = achromatopsia; H = hemianopia; nas = nasal;

lat = lateral; dor = dorsal; front = frontal; occ = occipital; Q = quadrantanopia; NR = not reported.

par = parietal; temp = temporal; wm = white matter; NR = not Agraphia: Lists whether deﬁcits in writing are reported in

reported. the original papers. yes = writing impaired; no = writing intact;

Speciﬁc abbreviations (in alphabetical order): AnG = angular slight = minor writing problems; NR = writing scores not reported;

gyrus; Cb = cerebellum; cc = corpus callosum; ccs = calcarine surface = surface agraphia.

sulcus/cortex; Cun = cuneus; fmj = forceps major; FrP = frontal Aphasia: Lists whether language deﬁcits are reported in the

pole; FuG = fusiform gyrus; Hi = Hippocampus; ic = internal original papers. Again methods and sensitiviy of assessment dif-

capsule; ITG = inferior temporal gyrus; LgG = lingual gyrus; fer widely between studies, and where no deﬁcits are reported,

LG = lateral geniculate nucleus; LOTG = lateral occipito-temporal some deﬁcits may have been present but not assessed. Abbre-

gyrus; Md = medulla; MOTG = middle occipito-temporal gyrus; viations: an = anomia; nam = naming deﬁcit; nam(RT) = elevated

MTG = middle temporal gyrus; O2 = middle/lateral occipital RTs in naming; wfd = word ﬁnding deﬁcit; dys = dysnomia;

gyrus; O3 = inferior occipital gyrus; O4 = fourth occipital gyrus comp = comprehension deﬁcit; Optic aph. = optic aphasia; NR = not

(posterior fusiform gyrus); OcP = occipital pole; or = optic reported.

R. Starrfelt, M. Behrmann / Neuropsychologia 49 (2011) 2283–2298 2297

References Friedman, R. B., & Hadley, J. A. (1992). Letter-by-letter surface alexia. Cognitive Neu-

ropsychology, 9, 185–208.

Gaillard, R., Naccache, L., Pinel, P., Clemenceau, S., Volle, E., Hasboun, D., et al. (2006).

Adair, J. C., Cooke, N., & Jankovic, J. (2007). Alexia without agraphia in Creutzfeldt-

Direct intracranial, FMRI, and lesion evidence for the causal role of left infer-

Jakob disease. Journal of the Neurological Sciences, 263, 208–210.

otemporal cortex in reading. Neuron, 50, 191–204.

Ajax, E. T., Schenkenberg, T., & Kosteljanetz, M. (1977). Alexia without agraphia and

Gelman, R., & Butterworth, B. (2005). Number and language: How are they related?

the inferior splenium. Neurology, 27, 685–688.

Trends in Cognitive Sciences, 9, 6–10.

Armbruster, A. P., & Wijdicks, E. F. (2006). Pure alexia: Past and present. Mayo Clinic

Geschwind, N., & Fusillo, M. (1966). Color-naming defects in association with alexia.

Proceedings, 81, 397.

Archives of Neurology, 15, 137–146.

Behrmann, M., Nelson, J., & Sekuler, E. B. (1998). Visual complexity in letter-by-letter

Goldstein, M. N., Joynt, R. J., & Goldblatt, D. (1971). Word blindness with intact central

reading: “Pure” alexia is not pure. Neuropsychologia, 36, 1115–1132.

visual ﬁelds. A case report. Neurology, 21, 873–876.

Behrmann, M., Plaut, D. C., & Nelson, J. (1998). A literature review and new data

Goodglass, H., Lindﬁeld, K. C., & Alexander, M. P. (2000). Semantic capacities of

supporting an interactive account of letter-by-letter reading. Cognitive Neu-

the right hemisphere as seen in two cases of pure word blindness. Journal of

ropsychology, 15, 7–51.

Psycholinguistic Research, 29, 399–422.

Benson, D. F., Brown, J., & Tomlinson, E. B. (1971). Varieties of alexia, Word and letter

Greenblatt, S. H. (1973). Alexia without agraphia or hemianopsia, Anatomical anal-

blindness. Neurology, 21, 951–957.

ysis of an autopsied case. Brain, 96, 307–316.

Binder, J. R., & Mohr, J. P. (1992). The topography of callosal reading pathways. A

Greenblatt, S. H. (1976). Subangular alexia without agraphia or hemianopsia. Brain

casecontrol analysis. Brain, 115, 1807–1826.

and Language, 3, 229–245.

Bub, D. N., Arguin, M., & Lecours, A. R. (1993). Jules Dejerine and his interpretation

Grossi, D., Fragassi, N. A., Orsini, A., De Falco, F. A., & Sepe, O. (1984). Residual reading

of pure alexia. Brain and Language, 45, 531–559.

capability in a patient with alexia without agraphia. Brain and Language, 23,

Bundesen, C. (1990). A theory of visual attention. Psychological Review, 97, 523–547.

337–348.

Buxbaum, L. J., & Coslett, H. B. (1996). Deep dyslexic phenomena in a letter-by-letter

Hasson, U., Levy, I., Behrmann, M., Hendler, T., & Malach, R. (2002). Eccentricity bias

reader. Brain and Language, 54, 136–167.

as an organizing principle for human high-order object areas. Neuron, 34(3),

Caffarra, P. (1987). Alexia without agraphia or hemianopia. European Neurology, 27,

65–71. 479–490.

Henderson, V. W. (1987). Is number reading selectively spared in pure alexia. Journal

Caplan, L. R., & Hedley Whyte, T. (1974). Cuing and memory dysfunction in alexia

of Clinical and Experimental Neuropsychology, 9, 41.

without agraphia. A case report. Brain, 97, 251–262.

Henderson, V. W., Friedman, R. B., Teng, E. L., & Weiner, J. M. (1985). Left hemi-

Celebisoy, N., Sagduyu, A., & Atac, C. (2005). Alexia without agraphia following

sphere pathways in reading: Inferences from pure alexia without hemianopia.

cerebral venous thrombosis associated with protein C and protein S deﬁciency.

Neurology, 35, 962–968.

Clinical Neurology and Neurosurgery, 108, 52–55.

Holtzman, R. N., Rudel, R. G., & Goldensohn, E. S. (1978). Paroxysmal alexia. Cortex,

Chialant, D., & Caramazza, A. (1998). Perceptual and lexical factors in a case of letter-

14, 592–603.

by-letter reading. Cognitive Neuropsychology, 15, 167–201.

Horikoshi, T., Asari, Y., Watanbe, A., Nagaseki, Y., Nukui, H., Sasaki, H., et al. (1997).

Cohen, L., & Dehaene, S. (1995). Number processing in pure alexia: The effect of

Music alexia in a patient with mild pure alexia: Disturbed visual perception of

hemispheric asymmetries and task demands. Neurocase, 1, 121–137.

nonverbal meaningful ﬁgures. Cortex, 33, 187–194.

Cohen, L., & Dehaene, S. (2000). Calculating without reading: Unsuspected residual

Imtiaz, K. E., Nirodi, G., & Khaleeli, A. A. (2001). Alexia without agraphia: A century

abilities in pure alexia. Cognitive Neuropsychology, 17, 563–583.

later. International Journal of Clinical Practice, 55, 225–226.

Cohen, L., Martinaud, O., Lemer, C., Lehericy, S., Samson, Y., Obadia, M., et al. (2003).

Ingles, J. L., & Eskes, G. A. (2008). A comparison of letter and digit processing in

Visual word recognition in the left and right hemispheres: Anatomical and func-

letter-by-letter reading. Journal of the International Neuropsychological Society,

tional correlates of peripheral alexias. Cerebral Cortex, 13, 1313–1333.

14, 164–173.

Cohen, L., Henry, C., Dehaene, S., Martinaud, O., Lehéricy, S., Lemer, C., et al.

Iragui, V. J., & Kritchevsky, M. (1991). Alexia without agraphia or hemianopia in pari-

(2004). The pathophysiology of letter-by-letter reading. Neuropsychologia, 42,

1768–1780. etal infarction. Journal of Neurology, Neurosurgery and Psychiatry, 54, 841–842.

Joy, H. H. (1947). Agnostic alexia without agraphia following trauma. American Jour-

Collignon, R. (1972). Case of Pure Alexia. International Journal of Mental Health, 1,

8–11. nal of Ophthalmology, 31, 811–820.

Judd, T., Gardner, H., & Geschwind, N. (1983). Alexia without agraphia in a composer.

Coslett, H. B., & Saffran, E. M. (1989). Evidence for preserved reading in ‘pure alexia’.

Brain, 106, 435–457.

Brain, 112, 327–359.

Karanth, P. (1981). Pure alexia in a Kannada-English bilingual. Cortex, 17, 187–198.

Coslett, H. B., Saffran, E. M., Greenbaum, S., & Schwartz, H. (1993). Reading in pure

Katz, R. B. (1990). Cross-modality word matching in letter-by-letter readers. Cortex,

alexia. The effect of strategy. Brain, 116, 21–37.

26, 65–76.

Crawford, J. R., & Garthwaite, P. H. (2005). Testing for suspected impairments and dis-

Kinsbourne, M., & Warrington, E. K. (1963). The localizing signiﬁcance of limited

sociations in single-case studies in neuropsychology: Evaluation of alternatives

simultaneous visual form perception. Brain, 86, 697–702.

using monte carlo simulations and revised tests for dissociations. Neuropsychol-

Lambon Ralph, M. A., Hesketh, A., & Sage, K. (2004). Implicit recognition in pure

ogy, 19, 318–331.

alexia: The Saffran effect—A tale of two systems or two procedures? Cognitive

Crawford, J. R., & Garthwaite, P. H. (2007). Comparison of a single case to a control or

Neuropsychology, 21.

normative sample in neuropsychology: Development of a Bayesian approach.

Landis, T., Regard, M., & Serrat, A. (1980). Iconic reading in a case of alexia without

Cognitive Neuropsychology, 24, 343–372.

agraphia caused by a brain tumor: A tachistoscopic study. Brain and Language,

Crawford, J. R., Garthwaite, P. H., & Gray, C. D. (2003). Wanted: Fully operational

11, 45–53.

deﬁnitions of dissociations in single-case studies. Cortex, 39, 357–370.

Lang, C. (1985). Palinoptic phenomena as an error mechanism in resolving alexia

Crawford, J. R., & Howell, D. C. (1998). Comparing an individual’s test score against

without agraphia—A case-report. European Neurology, 24, 248–253.

norms derived from small samples. The Clinical Neuropsychologist, 12, 482–

486. Lanzinger, S., Weder, B., Oettli, R., & Fretz, C. (1999). Neuroimaging ﬁndings in a

patient recovering from global alexia to spelling dyslexia. Journal of Neuroimag-

Cumming, W. J., Hurwitz, L. J., & Perl, N. T. (1970). A study of patient who had alexia

ing, 9, 48–51.

without agraphia. Journal of Neurology, Neurosurgery and Psychiatry, 33, 34–39.

Larsen, J., Baynes, K., & Swick, D. (2004). Right hemisphere reading mechanisms in

Dalmas, J. F., & Dansilio, S. (2000). Visuographemic alexia: A new form of a peripheral

a global alexic patient. Neuropsychologia, 42, 1459–1476.

acquired dyslexia. Brain and Language, 75, 1–16.

Lee, K. Y. (1966). Alexia without agraphia. American Journal of Ophthalmology, 62,

Daniel, M. S., Bolter, J. F., & Long, C. J. (1992). Remediation of alexia without agraphia:

1149.

A case study. Brain Injury, 6, 529–542.

Leegaard, O. F., Riis, J. O., & Andersen, G. (1988). Pure alexia” without hemianopia or

Dehaene, S. (2009). Reading in the brain. New York: Viking Penguin.

colour anomia. Acta Neurologica Scandinavica, 78, 501–505.

Dejerine, J. (1892). Contribution à l’étude anatomo-pathologique et clinique des dif-

Leff, A. P., Crewes, H., Plant, G. T., Scott, S. K., Kennard, C., & Wise, R. J. (2001). The

férentes variétés cécité verbale. Comptes Rendus Hebdomaladaires des Séances et

functional anatomy of single-word reading in patients with hemianopic and

Mémoires de la Société de Biologie, 4, 61–90.

pure alexia. Brain, 124, 510–521.

Di Pace, E., Guariglia, C., Judica, A., Spinelli, D., & Zoccolotti, P. (1995). Selective read-

Leff, A. P., Spitsyna, G., Plant, G. T., & Wise, R. J. (2006). Structural anatomy of pure

ing slowness in a traumatic patient with impairment in basic visual processes.

and hemianopic alexia. Journal of Neurology, Neurosurgery and Psychiatry, 77,

Journal of Clinical and Experimental Neuropsychology, 17, 878–899.

1004–1007.

Dogulu, C. F., Kansu, T., & Karabudak, R. (1996). Alexia without agraphia in multiple

Levin, H. S., & Rose, J. E. (1979). Alexia without agraphia in a musician after tran-

sclerosis [letter]. Journal of Neurology, Neurosurgery and Psychiatry, 61, 528.

scallosal removal of a left intraventricular meningioma. Neurosurgery, 4, 168–

Dufﬁeld, J. S., Desilva, R. N., & Grant, R. (1994). Pure alexia without agraphia—A

174.

classical cortical syndrome revisited. Scottish Medical Journal, 39, 178–179.

Lewis, J. P. (1995). Fast template matching. Vision Interface,

Erdem, S., & Kansu, T. (1995). Alexia without either agraphia or hemianopia in

Luhdorf, K., & Paulson, O. B. (1977). Does alexia without agraphia always include

temporal-lobe lesion due to herpes-simplex encephalitis. Journal of Neuro-

hemianopsia? Acta Neurologica Scandinavica, 55, 323–329.

Ophthalmology, 15, 102–104.

Maher, L. M., Clayton, M. C., Barrett, A. M., Schober Peterson, D., & Gonzalez

Farah, M. J., & Wallace, M. A. (1991). Pure alexia as a visual impairment: A reconsid-

Rothi, L. J. (1998). Rehabilitation of a case of pure alexia: Exploiting resid-

eration. Cognitive Neuropsychology, 8, 313–334.

ual abilities. Journal of the International Neuropsychological Society, 4, 636–

Fincham, R. W., Nibbelink, D. W., & Aschenbrener, C. A. (1975). Alexia with left

647.

homonymous hemianopia without agraphia, A case report with autopsy ﬁnd-

Mao-Draayer, Y., & Panitch, H. (2004). Alexia without agraphia in multiple sclerosis:

ings. Neurology, 25, 1164–1168.

Case report with magnetic resonance imaging localization. Multiple Sclerosis, 10,

Fiset, D., Gosselin, F., Blais, C., & Arguin, M. (2006). Inducing letter-by-letter dyslexia

705–707.

in normal readers. Journal of Cognitive Neuroscience, 18, 1466–1476.

2298 R. Starrfelt, M. Behrmann / Neuropsychologia 49 (2011) 2283–2298

Marangolo, P., Nasti, M., & Zorzi, M. (2004). Selective impairment for reading num- Rosazza, C., Appollonio, I., Isella, V., & Shallice, T. (2007). Qualitatively different forms

bers and number words: A single case study. Neuropsychologia, 42, 997–1006. of pure alexia. Cognitive Neuropsychology, 24, 393–418.

Marks, R. L., & Devito, T. (1987). Alexia without agraphia and associated Rosenﬁeld, I. (1988). The invention of memory: A new view of the brain (1 ed.). New

disorders—Importance of recognition in the rehabilitation setting. Archives of York: Basic Books.

Physical Medicine and Rehabilitation, 68, 239–243. Sekuler, E. B., & Behrmann, M. (1996). Perceptual cues in pure alexia. Cognitive

Miglis, M. G., & Levine, D. N. (2010). Intracranial venous thrombosis after placement Neuropsychology, 13, 941–974.

of a lumbar drain. Neurocritical Care, 12, 83–87. Shallice, T. (1988). From neuropsychology to mental structure (1 ed.). Cambridge:

Miozzo, M., & Caramazza, A. (1998). Varieties of pure alexia: The case of failure to Cambridge University Press.

access graphemic representations. Cognitive Neuropsychology, 15, 203–238. Shipkin, P. M., Gray, B. S., Daroff, R. B., & Glaser, J. S. (1981). Alexia without agraphia

Mycroft, R. H., Behrmann, M., & Kay, J. (2009). Visuoperceptual deﬁcits in letter-by- in a right-handed patient with a left ﬁeld defect and diffuse cerebral atrophy.

letter reading? Neuropsychologia, 47, 1733–1744. Neuro-Ophthalmology, 2, 123–128.

Patterson, K., & Kay, J. (1982). Letter-by-letter reading: Psychological descriptions of Starrfelt, R., & Gerlach, C. (2007). The visual what for area: Words and pictures in

a neurological syndrome. Quarterly Journal of Experimental Psychology A, Human the left fusiform gyrus. NeuroImage, 35, 334–342.

Experimental Psychology, 34, 411–441. Starrfelt, R., Habekost, T., & Leff, A. P. (2009). Too little, too late: Reduced visual span

Pena Casanova, J., Roig Rovira, T., Bermudez, A., & Tolosa Sarro, E. (1985). Optic and speed characterize pure alexia. Cerebral Cortex, 19, 2880–2890.

aphasia, optic apraxia, and loss of dreaming. Brain and Language, 26, 63–71. Starrfelt, R., Habekost, T., & Gerlach, C. (2010). Visual processing in pure alexia: A

Perri, R., Bartolomeo, P., & Silveri, M. C. (1996). Letter dyslexia in a letter-by-letter case study. Cortex, 46, 242–255.

reader. Brain and Language, 53, 390–407. Verma, A., Singh, N. N., & Misra, S. (2004). Transitory alexia without agraphia: A

Pillon, B., Bakchine, S., & Lhermitte, F. (1987). Alexia without agraphia in a disconnection syndrome due to neurocysticercosis. Neurology India, 52, 378–

left-handed patient with a right occipital lesion. Archives of Neurology, 44, 379.

1257–1262. Vincent, F. M., Sadowsky, C. H., Saunders, R. L., & Reeves, A. G. (1977). Alexia with-

Piras, F., & Marangolo, P. (2009). Word and number reading in the brain: Evi- out agraphia, hemianopia, or color-naming defect: A disconnection syndrome.

dence from a Voxel-based Lesion-symptom Mapping study. Neuropsychologia, Neurology, 27, 689–691.

47, 1944–1953. Warrington, E. K., & Langdon, D. (1994). Spelling dyslexia: A deﬁcit of the

Plaut, D. C., & Behrmann, M. (In press). Complementary neural representations for visual word-form. Journal of Neurology, Neurosurgery and Psychiatry, 57, 211–

faces and words: A computational exploration. Cognitive Neuropsychology. 216.

Price, C. J., & Humphreys, G. W. (1992). Letter-by-letter reading? Functional deﬁcits Warrington, E. K., & Langdon, D. W. (2002). Does the spelling dyslexic read by

and compensatory strategies. Cognitive Neuropsychology, 9, 427–457. recognizing orally spelled words? An investigation of a letter-by-letter reader.

Regard, M., Landis, T., & Hess, K. (1985). Preserved stenography reading in a patient Neurocase, 8, 210–218.

with pure alexia, A model for dissociated reading processes. Archives of Neurol- Warrington, E. K., & Shallice, T. (1980). Word-form dyslexia. Brain, 103, 99–112.

ogy, 42, 400–402. Wechsler, A. F., Weinstein, E. A., & Antin, S. P. (1972). Alexia without agraphia, A

Roberts, D. J., Lambon Ralph, M. A., & Woollams, A. M. (2010). When does less clinical and radiographical study of three unusual cases. Bulletin of the Los Angeles

yield more? The impact of severity upon implicit recognition in pure alexia. Neurological Society, 37, 1–11.

Neuropsychologia, 48(9), 2437–2446. Winkelman, M. D., & Glasson, C. C. (1984). Unilateral right cerebral representation

Rosati, G., Debastiani, P., Aiello, I., & Agnetti, V. (1984). Alexia without agraphia—A of reading in a familial left-hander. Neuropsychologia, 22, 621–626.

study of a case of verbal alexia without accompanying color-naming defect. Woods, B. T., & Poppel, E. (1974). Effect of print size on reading time in a patient

Journal of Neurology, 231, 20–25. with verbal alexia. Neuropsychologia, 12, 31–41.